Deadlock II

Tom Kelliher, CS 311

Apr. 2, 2012

Administrivia

Announcements

Exam in one week.

Assignment

From Last Time

Deadlock I.

Outline

1. Deadlock detection and recovery.

2. Deadlock avoidance.

3. Summary.

Coming Up

Linux kernel modules.

Deadlock Detection and Recovery

1. Permit deadlocks to occur, subsequently recover from them.

2. Periodically run deadlock detection routines.

3. Choose victim processes.

Deadlock Detection

Algorithmic deadlock detection:

1. $n$ processes.

2. $m$ resource types.

3. Available -- vector of length $m$.

4. Allocation -- matrix of size $n$ by $m$. Rows: processes; columns: resources.

5. Request -- matrix of size $n$ by $m$.

6. Finish -- vector of length $n$.

7. Work -- vector of length $m$.

The algorithm:

work = available;

for (i = 0; i < n; ++i)
   finish[i] = false;

while there is an i such that finish[i] == false 
and request[i] <= work
{
   finish[i] = true;
   work = work + allocation[i];
}

1. Running time?

2. How often should this be run? What are the tradeoffs?

3. Any processes for which finish is false are deadlocked.

Using a resource allocation graph to detect deadlock (5 processes, 3 resource types):

$${\rm Available} = \left[ \begin{array}{rrr} 0 & 0 & 0 \end{array} \right]$$

$${\rm Allocation} = \left[ \begin{array}{rrr} 0 & 1 & 0 \\ ... ... 3 & 0 & 3 \\ 2 & 1 & 1 \\ 0 & 0 & 2 \end{array} \right]$$

$${\rm Request} = \left[ \begin{array}{rrr} 0 & 0 & 0 \\ 2 ... ... 0 & 0 & 0 \\ 1 & 0 & 0 \\ 0 & 0 & 2 \end{array} \right]$$

Retry with this request matrix:

$${\rm Request} = \left[ \begin{array}{rrr} 0 & 0 & 0 \\ 2 ... ... 0 & 0 & 1 \\ 1 & 0 & 0 \\ 0 & 0 & 2 \end{array} \right]$$

Deadlock Recovery

Some recovery mechanisms:

1. Terminate all deadlocked processes. What about threads (say only some of the threads in a task are deadlocked)?

2. Terminate processes one at a time.
   1. How do you choose the victim?

   2. When do you stop terminating victims?

3. ``Rolling back'' a process and preempting resources.
   1. Process termination is drastic and messy.

   2. Checkpoints.

   3. How much state has to be checkpointed?

   4. Starvation.

Deadlock Avoidance

Always maintain the system in a safe state:

Grant a resource request only if resulting state is safe.

Safe state:

1. There exists a sequence in which all resource requests can be satisfied.

2. The operating system is in control of the resource situation.

Unsafe state:

1. The operating system is no longer in control of the resource situation.

2. Processes are in control and can cause a deadlock.

Banker's algorithm used to maintain safe state.

Additional matrix required: Claim -- maximum number of each resource type needed by each process.

For each resource request which can be satisfied
{
   simulate:
      satisfy the request;
      update state matrices;
      turn all remaining claims into requests;
      test for deadlock;

   if simulation resulted in deadlock
      defer the request;
   else
      grant the request;
}

Example

Assume a maximum-claim serially reusable system with four processes and three resource types. The claim matrix is given by

$$C = \left[ \begin{array}{rrr} 4 & 1 & 4 \\ 3 & 1 & 4 \\ 5 & 7 & 13 \\ 1 & 1 & 6 \end{array}\right],$$

where $C_{i,j}$ denotes the maximum claim of process $i$ for resource $j$. The total units of each resource type are given by the vector $(5~8~16)$. The allocation of resources is given by the matrix

$$A = \left[ \begin{array}{rrr} 0 & 1 & 4 \\ 2 & 0 & 1 \\ 1 & 2 & 1 \\ 1 & 0 & 3 \end{array}\right],$$

where $A_{i,j}$ denotes the number of units of resource $j$ that are allocated to process $i$.

1. Determine if the current state of the system is safe.

2. Determine if granting a request by process 1 for 1 unit of resource 1 can be safely granted.

3. Determine if granting a request by process 3 for 6 units of resource 3 can be safely granted.

Summary

1. Some resources are easy to keep out of deadlock: CPU cycles, memory.

2. No one ``silver bullet'' -- must combine mechanisms.

3. Must tradeoff between cost of protection and cost of deadlock.

Tom Kelliher